The single-crystal X-ray analysis of trigonal C-C-A-T-T-A-A-T-G-G, and its comparison with orthorhombic C-G-A-T-T-A-A-T-C-G, have shown that the A-T-T-A-A-T sequence has limited polymorphism under the influence of packing forces from neighboring molecules in the crystal. The T-A step is intrinsically variable. It is not inconsistent with a large propeller twist, a narrow minor groove, and a single spine of hydration, as has sometimes been claimed on theoretical grounds. The T-A step does show a persistent positive roll, in a direction that compresses the major groove, and this may be a significant factor in macroscopic DNA curvature induced by phased A-tracts. A-tracts, as understood in this paper, include A-A and A-T steps, but not the T-A step, which is disruptive. Three conclusions regarding A-tract-induced curvature can be drawn from this and other X-ray crystal structure analyses, and from key gel retardation experiments: (1) The A-tract bending model is disqualified on two grounds: (i) tilt-wedge bending within A-tracts is incompatible with the observed direction of curvature; (ii) roll-wedge bending within A-tracts is contradicted by every crystal structure analysis, and is inconsistent with gel retardation results for (G-C-A-A-A-A-T-T-T-T)n and for (A-A-A-A-A-T-T-T-T-T)n. (2) The junction bend model is contradicted by crystallography because: (i) the inclination of base-pairs does not change between A-tract and non-A-tract regions of helix; and (ii) the observed bends at GC/AT junctions are roll-wedge bends, not tilt-wedge as the junction bend model demands. (3) The non-A-tract bending model is consistent with both gel retardation data and with X-ray crystallography, and must be regarded as the only consistent model for A-tract bending.